A Multiscale Data-Driven Forecasting Framework for Optimum Field Development Planning

Authors - Amir Salehi, Gill Hetz, Soheil Esmaeilzadeh, Feyisayo Olalotiti, and David Castineira

Download the Paper

Abstract - The process of identifying and performance forecasting of the remaining, feasible, and actionable field development opportunities (FDOs) is the core component of optimum field development planning and management. In the present work, we introduce a multiscale data-driven forecasting framework that applies a series of novel technologies to provide short- and long-term production forecast and optimization for both field and well level performance. Our workflow can be applied to a comprehensive FDO inventory including behind-pipe recompletion, infill drilling, and sidetrack opportunities. Using smart spatio-temporal clustering, we automatically divide the reservoir into a specific number of compartments with distinct static and dynamic properties, in which the direction-dependent multiphase flow communication is a function of nonlocal phase potential differences. The reservoir connectivity structure is encoded in an adjacency matrix describing the neighbor and non-neighbor connections of comprising compartments. We then apply a recently proposed robust Ensemble Smoother Levenberg-Marquardt (rES-LM) method to generate plausible model realizations which replicate the reservoir energy by adjusting first-order model parameters such as pore volumes, fault transmissibilities, aquifer strength, and matrix-fracture split. These calibrated upscaled network models serve as pre-conditioner for a detailed model calibration step. We carry out a second round of full-scale reservoir simulation model calibration, anchoring updates on large-scale model parameters estimated from the network model. Representative models are further improved in a sensitivity-based local inversion step to match multiphase production data at the well-level. This structured, multiscale approach offers improved stability in reservoir model calibration. Finally, calibrated models are directly passed to the forecasting and optimization engine to assess and optimize field opportunities and development scenarios. The proposed workflow is applied to a major fractured offshore field in South America. Leveraging the fast forward model, an efficient ensemble-based history matching framework was applied to reduce the uncertainty of the global reservoir parameters, such as inter-blocks and aquifer-reservoir communications, and fault transmissibilities. The ensemble of history-matched models was then used to provide a probabilistic forecast and optimization for different field development scenarios. A novel hybrid approach is presented in which we couple a physics-based nonlocal modeling framework with data-driven clustering techniques to provide a fast and accurate multiscale modeling of compartmentalized reservoirs. Our approach facilitates a flexible framework to rapidly generate reliable forecasts and quantify associated uncertainties in a robust manner. This advantage in flexibility and robustness is tied to our fast and automated two-stage model calibration workflow that leads to substantial saving in computational time. This research also adds to the literature by presenting a comprehensive work on spatio-temporal clustering for reservoir studies’ applications that consider the clustering complexities, the intrinsic sparse and noisy nature of the data, and the interpretability of the outcome.

Keywords: Super-Resolution, PDEs, Physics-Constrained, Deep Neural Networks